Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Regulatable AI Systems: Technical Gaps and Policy Opportunities

Xudong Shen, Hannah Brown, Jiashu Tao, Martin Strobel, Yao Tong, Akshay Narayan, Harold Soh, Finale Doshi-Velez

TL;DR

The paper investigates how AI systems can be made regulatable by examining two public-sector procurement checklists (CDADM and WEF) and mapping their technical criteria into data checks, monitoring, explanations, design, privacy, and human-in-the-loop. It assesses what is currently feasible and identifies gaps requiring new AI innovations or interdisciplinary collaboration to enable verifiable regulatory compliance. The authors outline concrete research directions across data governance, post-hoc monitoring, global and local explanations, objective design, privacy, and human–AI interaction. The work aims to bridge policy aspirations and technical capabilities, providing a framework to advance regulatable AI with practical pathways for governance and oversight.

Abstract

There is increasing attention being given to how to regulate AI systems. As governing bodies grapple with what values to encapsulate into regulation, we consider the technical half of the question: To what extent can AI experts vet an AI system for adherence to regulatory requirements? We investigate this question through the lens of two public sector procurement checklists, identifying what we can do now, what should be possible with technical innovation, and what requirements need a more interdisciplinary approach.

Towards Regulatable AI Systems: Technical Gaps and Policy Opportunities

TL;DR

The paper investigates how AI systems can be made regulatable by examining two public-sector procurement checklists (CDADM and WEF) and mapping their technical criteria into data checks, monitoring, explanations, design, privacy, and human-in-the-loop. It assesses what is currently feasible and identifies gaps requiring new AI innovations or interdisciplinary collaboration to enable verifiable regulatory compliance. The authors outline concrete research directions across data governance, post-hoc monitoring, global and local explanations, objective design, privacy, and human–AI interaction. The work aims to bridge policy aspirations and technical capabilities, providing a framework to advance regulatable AI with practical pathways for governance and oversight.

Abstract

There is increasing attention being given to how to regulate AI systems. As governing bodies grapple with what values to encapsulate into regulation, we consider the technical half of the question: To what extent can AI experts vet an AI system for adherence to regulatory requirements? We investigate this question through the lens of two public sector procurement checklists, identifying what we can do now, what should be possible with technical innovation, and what requirements need a more interdisciplinary approach.
Paper Structure (32 sections, 1 figure)

This paper contains 32 sections, 1 figure.

Table of Contents

  1. Introduction
  2. Public Sector AI Procurement Checklists
  3. Inputs of the Model: (Pre-training) Data Checks
  4. What we know how to do
  5. Directions requiring additional AI innovation
  6. Areas that require interdisciplinary engagement
  7. Outputs of the Model: (Post-hoc) System Monitoring
  8. What we know how to do
  9. Directions requiring additional AI innovation
  10. Areas that require interdisciplinary engagement
  11. Inspecting the Model: Global Explanations for Model Validation
  12. What we know how to do
  13. Directions requiring additional AI innovation
  14. Areas that require interdisciplinary engagement
  15. Inspecting the Model: Local Explanations about Individual Decisions
  16. ...and 17 more sections

Figures (1)

  • Figure 1: Connections between the main sections in this article and the high-level CDADM requirements and WEF guidelines. We exclude the WEF guidelines on strategy alignment, multidisciplinary teams, and fair business competition which do not directly motivate CS research.